This paper proposes a revised Glosten-Jagnnathan-Runkle (GJR) model for estimating hedge ratios. The model can take into account three important characteristics in the return behavior, i.e., fat-tailed distribution, leverage effect, and spot-futures spread. Hedge performance in terms of the White's (2000) reality check is conducted. Our results demonstrate that the generalized autoregressive conditional heteroskedasticity (GARCH) model that considers both fat-tailed distribution and asymmetric effects of the spread provides the best hedging effectiveness for longer horizons.